Developer: Vyom AI Base model: Qwen/Qwen3.5-4B (fine-tuned with LoRA) Persona: Indian AI with Hinglish personality — baked into weights Tagline: Infinite Intelligence. Refined.

No system prompt needed. VYOM identity is fine-tuned in.

About

VYOM (व्योम — meaning infinite space/sky) is a fine-tuned conversational AI built on Qwen3.5-4B via 16-bit LoRA. Persona, tone, and Hinglish fluency are baked directly into the model weights — no system prompt required at inference.

Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained("anilsuthar2004/VYOM_4B")
model = AutoModelForCausalLM.from_pretrained(
    "anilsuthar2004/VYOM_4B",
    torch_dtype=torch.float16,
    device_map="auto"
)

# No system prompt needed — persona is in the weights!
messages = [{"role": "user", "content": "Tu kaun hai?"}]
text = tokenizer.apply_chat_template(
    messages, tokenize=False,
    add_generation_prompt=True,
    enable_thinking=False
)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
out = model.generate(
    **inputs,
    max_new_tokens=200,
    temperature=0.7,
    top_p=0.8,
    top_k=20,
    repetition_penalty=1.0
)
print(tokenizer.decode(
    out[0][inputs["input_ids"].shape[1]:],
    skip_special_tokens=True
))

Training details

Field	Value
Method	LoRA (16-bit, not QLoRA)
LoRA rank	16
LoRA alpha	32
Epochs	5
Batch size	4 (2 per GPU × 2 T4s)
Learning rate	2e-4
Scheduler	cosine
Optimizer	adamw_8bit
Platform	Kaggle 2×T4
Seq length	2048
Training data	60 VYOM Q&A pairs (Hinglish)

Inference parameters

Parameter	Value
temperature	0.7
top_p	0.8
top_k	20
repetition_penalty	1.0